My application needs to verify if address in the input matches with the supporting document user has uploaded. Currently using AWS Rekognition to detect text and verify with user provided input. But this approach does not seem very reliable as documents can be easily edited and tampered with. What would be the best approach to verify text and authenticity (not tampered with) of a document image using Rekognition?
AWS Rekognition cannot verify if the image was tempered or not. You at best try to check if the text in the image is added in the proper position(X-Y co-ordinates) of the image. AWS Rekognition detects the text and its position and also gives the X-Y co-ordinates in the image. This can help reduce it tempered images but not completely. If you want to verify it completely then you should look for some third party tool.